Rightsizing
Last updated 2026-06-04
Rightsizing is the practice of matching the size of a cloud resource, such as a virtual machine, database, or container, to a workload's actual demand. Many resources are provisioned for peak or guessed-at capacity and then run far below it, paying for headroom they never use. Rightsizing analyzes utilization metrics, typically CPU, memory, IOPS, and network, often at the P95 percentile to preserve a safety margin, then recommends a smaller or more appropriate instance type or resource request. For example, a database on a large instance that rarely exceeds a fraction of its CPU and memory can be moved to a smaller class without affecting performance. Because the change follows observed behavior rather than guesswork, it is one of the highest-impact, lowest-risk cloud cost optimizations, and it should be revisited as usage patterns shift. LevelFour rightsizes resources from observed usage and delivers each change as a reviewable infrastructure-as-code pull request.
Frequently asked questions
- How is rightsizing different from autoscaling?
- Rightsizing sets the correct baseline size of a resource based on its observed demand, usually CPU, memory, IOPS, and network. Autoscaling adjusts how many instances run as load changes. They are complementary: rightsizing picks the right unit, autoscaling adjusts the count of those units automatically.
- What metrics should you use to rightsize a cloud resource?
- Rightsizing relies on real utilization data, typically CPU, memory, IOPS, and network throughput gathered over a representative period. Practitioners often size to the P95 percentile rather than the average, so the resource handles typical peaks with a safety margin while ignoring rare, brief spikes that would otherwise force permanent over-provisioning.
See also
LevelFour automates this across AWS, GCP, Azure, and Kubernetes with automated infrastructure-as-code pull requests.